Weight of evidence model based on GIS to evaluate landslides

Weight of evidence model based on GIS to evaluate landslides

WATER AND GEOSCIENCE
Weight of evidence model based on GIS to evaluate landslides susceptibility in
central Rif of Morocco
AHMED NASR-EDDINE EL FAHCHOUCH, LAHCEN AIT BRAHIM, MOHAMED MASTERE,
ABDLLAH ABDELOUAFI
Department of Earth sciences
University Mohammed V, Faculty of Sciences
Avenue Ibn Battuta Rabat - Agdal, PO Box 1014, Rabat,
MOROCCO
[email protected]
Abstract: - The aim of this study is to replace landslides in the specific geodynamic context of the central Rif. We
adopted a multisource approach that includs remote sensing data, geology, geotechnical characteristic, landform and
climatology. Then, we elaborated a database in a GIS on the permanent and triggering parameters (slope, lithology,
fracturing, hydrographical network, land use….
Weights-of-evidence (WOE) is a quantitative ‘data driven’ method used to combine datasets uses the log–linear form
of the Bayesian probability model to estimate the relative importance of evidence by statistical means, This method
has been carried out at Beni Ahmed maps at 1:50000 scale. LandSat+ image has been used to map landslides occurred
in the study area, to produce Land use map and to improve the faults coverage extracted initially from geological
maps. The derived data were integrated to a GIS with others parameters, derived form geologicaI maps (lithological,
thickness), topographical maps (slope gradient, aspect, hydrological network), and thematic maps.
An identification and inventory of more than 350 mass movements of large scale covering our study area, the most
abundant type, number and area occupied, is represented by landslides (36%) over an area of 46.269 Km2.
The obtained results are confronted with the reality of the land (map of landslides that we have done) to adapt the
model to the specificities of the study area.
Key-Words: Landslides; Weight of evidence; GIS; susceptibility; central Rif; spatial analysis;
1 Introduction:
Due to its special geographical position, the Rif’s chain
is the northernmost part of Morocco; it is characterized
by complex geological terrain, morphology in steep
terrain, dense precipitation.
The combination of factors pedo-geological, climatic,
topographical and anthropogenic made the Rif area
undoubtedly the most exposed to natural phenomena
such as landslides. The effects of these phenomena are
The landslides cover a wide variety of natural
phenomena. All shares a destabilized displacement of
materials. These complex phenomena can be point,
superficial, limited in space and time but also rapid and
Indeed, Morocco, the mapping of natural hazards has
started for the first time in the sixties by Avenard
(1965) who mapped erosion in the basin of Sebu. In
1968, Millies-Lacroix [4] sets for all the Rif card
predictive of landslides in 1/1000000ème [5]. Insofar as
it is often difficult to quantify a level of hazard,
frequently, only the susceptibility to instability field is
analyzed. Susceptibility expresses the probability that a spatial
type of phenomenon occurs in an area for different
local environmental conditions. This assumes that all
ISSN: 1790-5095
especially important when they affect areas inhabited
more or less vulnerable (depending on the number of
people, housing, infrastructure and economic
activities).
The BeniAhmed region is part of the Rif; it presents
several signs of instability. While some areas remain
relatively stable, others are subject to factors of
instability or a shift asset.
widespread affecting then slopes intact ([1]; [2]). They
may be active, latent, inactive, or potential. Some may
present a danger to human lives and are responsible for
damages large and costly ([3];[2]).
phenomena are identified and classified, and they recur
in the same geological, geomorphological, hydrological
and climatic phenomena known.
The evaluation of the susceptibility involves three steps
([6]; [7]; [8]; [9]; [2]): (a) The inventory phenomena of
landslides where each phenomenon is distinguished by
its type, (b) The parameter mapping field the most
significant (factors predisposition) for the case of
spatial phenomena and their analysis, (c) The definition
of relative weights to each factor blamed for the
localization of landslides.
180
ISBN: 978-960-474-160-1
WATER AND GEOSCIENCE
Fig. 1: Methodological approach advocated (evidence theory)
The definition of these weights reflects the spatial
relationship between the landslides and predisposing
factors and is defined by various methods.
The techniques of spatial analysis by GIS are
increasingly used to assess susceptibility to
landslides. One of these methods is weight of evidence
theory ([10]; [11]). However, the main drawback of this
technique is the problem of information redundancy
between predictive factors.
This work aims the following objectives: (a) diagnosis
of landslides occurred in the study area through their
inventories, their maps and their characterizations, (b)
the design of a GIS database in ArcGIS 9.2
has causative factors of the four types of landslides, (c)
Evaluation of conditional independence of these factors
(d) the production of the map of susceptibility to
landslides through a bivariate probabilistic model based
on the Bayes theorem (theory of evidence).
3 Methodology and strategy: Weight of
evidence (WOE):
Fig.1 presents the methodology used and data
entry. The data preparation, archiving and simulations
were performed in ArcGIS 9.2 GIS environment. An
indirect approach based on statistical models of spatial
analysis was used to quantify the susceptibility. The
susceptibility is defined as a space probability of
landsliade occurs in an area for different local
environmental conditions. The susceptibility has been
simulated by a model based on the theory of evidence.
Weight of evidence is a quantitative ‘data-driven’
method used to combine datasets. The method uses the
log-linear form of the Bayesian probability model to
estimate the relative importance of evidence by
statistical means. This method was first applied to
landslide susceptibility mapping by ([12]; [2]). Prior
probabilities (PriorP) and posterior probabilities (PostP)
are the most important concepts in the Bayesian
approach. PriorP is the probability that a TU (terrain
unit) contains the RV (response variable) before taking
PVs (predictive variables) into account, and its
estimation is based on the RV density for the study
area. This initial estimate can be modified by the
introduction of other evidences. PostP is then estimated
according to the RV density for each class of the PV.
The model is based on the calculation of positive W+
and negative W− weights, whose magnitude depends
on the observed association between the RV and the
PV.
2 Geomorphological setting:
The study area is located in the north of Morocco
(fig.2), it is highly affected by landslide hazards. The
hypsometric prints in the region an important character
mountainous, it is marked by the collection of valleys,
steep
slopes
and
rock
faces
fallout
in
clumps, Characterized by a mountain climate with a
Mediterranean influence. The topography is hilly,
which makes isolation and difficult access. This
mountainous region characterized by wet altitudes
relatively low, not exceeding 700 m and rises gradually
up to 1700 m north of the region, specifically in Jebel
Taria. The
morphology
shows
peaks
with
approximately the same altitude 700m, separated by
basins elongated towards the NE.
ISSN: 1790-5095
181
ISBN: 978-960-474-160-1
WATER AND GEOSCIENCE
posterior probability, which updates the prior
probability. When multiple predictors are combined,
areas that have a weight higher or lower respectively
correspond to a greater or smaller probability of finding
the landslides.
This statistical model is introduced in the ArcGIS 9.2
through the ArcSDM extension [14]. The model
calculates the prior probabilities, posterior probabilities
and hypothesis testing type X2. The procedure to
determine the best combination is made add one to one
every predictor.
The maps show a good agreement with the landslide
inventory map. The surfaces of high, moderate, low and
null susceptibility are 12 Km2, 257 Km2, 330 Km2,
and 26 Km2, something that is quite understandable
given the extent occupied by these sites. This leads us
to propose to incorporate other factors namely
hydrographical networks, seismicity and the anthropic
factor in its various aspects (construction of roads,
quarrying, earthworks, etc.).
The analysis of the susceptibility map developed (Fig.
2) shows that the high susceptibility (Strong possibility
of triggering landslides) is largely concentrated in
northern and west parts of the study area where local
environmental conditions are very favourable to the
trigging (combination of a slope gradient> 25 °, Relief
delivered, strong fracturing, extremely degraded soils
or dissected, the absence of forest and vegetation or
poorly maintained).
In Eqs. (1) and (2), B is a class of the PV and the
overbar sign ‘¯’ represents the absence of the class
and/or RV. The ratio of the probability of RV presence
to that of RV absence is called odds [10]. The WOE for
all PVs is combined using the natural logarithm of the
odds (logit), in order to estimate the conditional
probability of landslide occurrence. When several PVs
are combined, areas with high or low weights
correspond to high or low probabilities of presence of
the RV.
Mapping of susceptibility occurs in several steps, the
first corresponds to the statistical analysis of the
landslides observed (identification and inventory of
landslides), the second step is the characterization and
identification of factors affecting (parameters
predisposition)
lithology,
fracturing,
slope,
precipitation, etc.. The third is the evaluation of the
conditional independence of predictive variables. The
fourth is the application of the bivariate approach by
theory of evidence.
4 Results:
Different weights can be summed calculated using the
natural logarithm of odds called logit. In this case the
contrast C (C = W + - W-) gives a measure of spatial
association between the predictors and landslides
[13]. This contrast has a value of zero when these two
variables are completely independent. The contrast
value gives a first overview to accept or reject a
predictor in estimating the spatial correlation between
this and the landslides. Calculations of values of W +
and W-for all selected variables used to calculate the
Fig.2: (a) Distribution of landslides inventory;
(b) Susceptibility map produced with the weight of evidence model
ISSN: 1790-5095
182
ISBN: 978-960-474-160-1
WATER AND GEOSCIENCE
[3] Maquaire, 2002, Aléas géomorphologiques
(mouvements de terrain) : processus, fonctionnement,
cartographie.Mémoire d'habilitation à Diriger des
Recherches : Université Louis Pasteur, Strasbourg, 219
p. + 1 volume d'annexes.
5 Conclusions:
Mush of the landslide susceptibility work is based on
the assumption that “the past is key to the future”, and
that historical landslides and their causal relationships
can be used to predict future ones. As soon as there are
changes in the causal factors (e.g. a road with steep cuts
is constructed in a slope which was considered as low
susceptibility before) [15].
This study has demonstrated the necessity of using
specific procedures for landslide susceptibility
assessment by bivaraite methods, especially at 1 :50
000 scale in the Morocco Rif zone . The thematic maps
introduced in the model represent slope gradient,
lithology, land use, precipitation, fracturing.
The susceptibility maps describe the conditional
probability of future landslide occurrences, which
depends on the values of unstable landform densities, it
proves to be best suited to guide choices in
implementation of development sites, the level of urban
extensions, and that at the layout of new roads and
highways in the national development program in the
Northern provinces.
The maps are developed documents to assist decision
making for any development plan in Beni Ahmed area
(urban sprawl, villages, roads, bridges, factories...).
[4] Millies-Lacroix, 1968, les glissements de terrains.
Présentation d’une carte prévisionnelle des le domaine
de masse dans le rif (Maoc Septentrional): Mines et
géologie, n°27.p45-55
[7] Soeters and Van Westen, 1996, Slope instability,
recognition, analysis, and zonation, in Turner A.K.,
Schuster R.L. (eds), Landslides investigation and
mitigation, Transport Research Board, National
Research Council: special report 247, 1996, p.129-177.
[5] Sossey Alaoui, 2005, Traitement et intégration des
données satellitaires optiques et Radar dans un SIG en
vue de l’obtention de carte de l’aléa lié aux instabilités
de terrain dans la péninsule de Tanger (Rif
septentrional, Maroc): Doctorat, Université Mohamed
V, Faculté des Sciences, Rabat. 175p.
[12] Van Westen, C.J., (1993) : Application of
Geographic Information Systems to Landslide Hazard
Zonation. Ph-D Dissertation, Technical University
Delft: ITC-Publication Number 15, ITC, Enschede, The
Netherlands, 245 p.
[9] Van Westen, C.J., Van Asch, T.W.J., Soeters, R.,(
2006) : Landslide hazard and risk zonation: why is it
still sodifficult: Bulletin of Engineering Geology and
the Environment, 65, 167–184.
References:
[10] Bonham-Carter, 1994, Geographic Information
System
for
Geoscientists:
Modelling
with
GIS:Pergamon Press, Oxford, 398p.
[13] Yannick Thiery and al., 2005, Analyse spatiale de
la susceptibilité des versants aux mouvements de
terrain, comparaison de deux approches spatialisées par
SIG : Revue internationale de Géomatique/European
Journal of GIS and Spatial Analysis, pp. 227-245.
[6] Carrara and al. 1995, GIS technology in mapping
landslide hazard: Geographical Information Systems in
Assessing Natural Hazards, p. 135-176.
[11] Yannick Thiery, 2006, Test of Fuzzy Logic Rules
for Landslide Susceptibility Assessment, SAGEO 2006.
Colloque International de Géomatique et d'Analyse
Spatiale: Recherches & Développements. Strasbourg,
16 p.
[1] Flageollet, 1989, Les mouvements de terrain et leur
prévention: Masson, Paris. 218 p.
[14] Kemp L.D., Bonham-Carter G.F., Raines G.L. and
Looney C.G., 2001, Arc-SDM: ArcViewextension for
spatial data modelling using weights of evidence,
logistic regression, fuzzy logic and neural network
analysis, 2001, http://ntserv.gis.nrcan.gc.ca/sdm/ .
[2] Yannick Thiery and al., 2007, Susceptibilité du
Bassin de Barcelonnette (Alpes du sud, France) aux
‘mouvements
de
versant’:
cartographie
morphodynamique, analyse spatiale et modélisation
probabiliste: Doctorat de l'universite de caen. paris.
P440.
[8] Leroi, 1996, Landslide hazard - risk maps at
different scales: Objectives, tools and developments.
Proceedings of the Seventh International Symposium on
Landslides: Trondheim, Norway, June 17-21 , vol. 1, p.
35-51.
ISSN: 1790-5095
[15] Van Westen and al., 2008, spatial data for
landslide susceptibility, hazard, and vulnerability
assessment: An overview, Engineering Geology 102,
112-131.
183
ISBN: 978-960-474-160-1